Publication: Ego-centred models of social networks: the social atom
Loading...
Identifiers
Publication date
2019-05-30
Defense date
2019-05-30
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This thesis set out to contribute to the realm of social physics, with a particular
focus on human social networks. Our approach, however, is somewhat
di
erent from what is typical in disciplines such as complex systems or statistical
physics. Rather than simplifying the features of the constituents of
our system (people), and stressing their rules of interaction, we focus on
better understanding those very same constituents, modelling them as social
atoms. Our rationale is that a better understanding of such an atom
may shed light on how (and why) it interacts with other atoms to form
social collectives.
Given its robustness and the evolutionary roots of its premises, we use
the Social Brain Hypothesis as our departure point. This theory states that
the evolutionary drive behind the development of large brains in humans
was the need to process social information and that the limited capacity of
our brains imposes a limit to the number of relationships we can manage—
the so-called “Dunbar’s number”, roughly 150. Moreover, evidence keeps
revealing that these relationships are further organised in a series of hierarchically
inclusive layers with decreasing emotional intensity, whose sizes
exhibit a more or less constant scaling. Notwithstanding the empirical evidence,
neither the presence of scaling in the organisation of personal networks
nor its connection with limited cognitive skills had been explained
so far.
In Chapter 2 we present a mathematical model that solves this puzzle.
The assumptions of the model are quite simple, and well founded on empirical
evidence. Firstly, the number of relationships we maintain tends
to be stable on average. Secondly, these relationships are costly, and our resources are limited. With these two premises, our results show that the
hierarchical organisation emerges naturally from the principle of maximum
entropy. Not only that, but we also predict a hitherto unnoticed regime of
organisation whose existence we prove using several datasets from communities
of immigrants.
The former model considers that relationships can only belong to a
discrete set of categories (layers). In Chapter 3 we extend it so that relationships
are classified in a continuum. This modification allows us to test
the model with data from very di
erent sources such as online communications,
face-to-face contacts, and phone calls. Our results show that the two
regimes of organisation found in the previous model persist in this variant,
and reveal the underlying existence of a (universal) scaling parameter
which does not depend on any particular number of layers.
To incorporate these ideas into socio-centric models, we build on the
so-called Structural Balance Theory. This theory, underpinned by psychological
motivations, posits that the structure of social networks of positive
and negative relationships are highly interdependent. However, the theory
has received little empirical validation, and negative social relationships
are poorly understood—both from an ego-centric and a socio-centric perspective.
For that reason, we turn to developing an experimental software
in order to gather data within a school.
In Chapters 4 and 5 we present results from these experiments. In
Chapter 4 we analyse the socio-centric networks using machine learning
techniques and find that the structure of positive and negative networks
is indeed very much connected. Besides, we study the two types of networks
separately, showing that they exhibit quite distinct features and that
gender e
ects in negative social networks are weak and asymmetrical for
boys and girls. In Chapter 5, on the other hand, we focus on the structure
of negative personal networks. Remarkably, using data from two di
erent
experimental settings, we show that the structure of personal networks
of negative relationships mirrors that of the positive ones and exhibits a
similar scaling—albeit their size is significantly smaller.
Chapter 6 summarises our results and presents future (and current) lines
of investigation. Among them, we outline a model of a social fluid that
uses the insights gained with this thesis to build a model of social collectives
as ensembles of personal networks. This model is compatible, at the micro-level, with the observations of the social brain hypothesis, and, at
the macro-level, with the premises of the structural balance theory.
Description
Mención Internacional en el título de doctor
Keywords
Complex systems, Structural Balance Theory, Machine learning techniques, Social atom, Quantitative sociology, Social networks, Behaviuor models